limma-voom model design
1
0
Entering edit mode
9 months ago

Would there be a difference between testing a condition within an interaction term or on its own? For example, if the dataset was the same for both of these models and we wanted to look at condition only, would we expect the number of differentially expressed genes to be the same for both models?

model.matrix(~participant + time*condition)

vs

model.matrix(~participant + condition)
differential-gene-expression limma-voom limma • 379 views
ADD COMMENT
0
Entering edit mode
9 months ago
LauferVA 4.2k

michaelhojungyoon , first a bit of pedantry if you don't mind - this question is meaningless in the absence of a model matrix. For instance, if we were to look at your data and realize that time and condition are perfectly collinear, then for the purposes of your question, the answer would be no.

in general, though, they do differ, and you absolutely need to understand this to "get" biology.

suppose you have two ways of making glucose, pathway A and pathway B.

if you knock out pathway A, nothing happens because pathway B can easily meet the demands of the cell even without any contribution from pathway A.

likewise, if you knock out pathway B, nothing happens because pathway A can also easily meet the demands of the cell even without any contribution from pathway B.

So, let's say you make a model, and you look at the main effect of knocking out pathway A, then the main effect of knocking out pathway B. you'd find in both cases that there was no effect.

But, now let's knockout both at the same time. here, you'd find that the double KOs have very low glucose production, while the pathway A KO and pathway B KO's are fine, and the double WT is fine.

thus, what you are saying is that it is the interaction of the pathway A KO and pathway B KO that produces an effect, not either alone, etc.

however, this entirely depends on the biological question -in other cases, this perhaps knocking out pathway A would reduce level to 50%, same with pathway B - in this case an interaction effect might be weak or absent, beacuse the effect is detectable using only one bit of information or the other..

ADD COMMENT

Login before adding your answer.

Traffic: 2565 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6